deepccc 0.1.16 → 0.1.18

package/dist/context.d.ts

@@ -52,7 +52,7 @@ export interface BuiltinContextOptions {
     keepRecentMessages?: number;
 }
 export declare const DEFAULT_BUILTIN_CONTEXT_DIR: string;
-export declare const DEFAULT_COMPACT_AT_TOKENS = 48000;
+export declare const DEFAULT_COMPACT_AT_TOKENS = 128000;
 export declare const DEFAULT_KEEP_RECENT_MESSAGES = 16;
 export declare function normalizeBuiltinSessionId(value: string): string;
 export declare function defaultBuiltinSessionId(cwd?: string): string;

package/dist/context.js

@@ -3,10 +3,10 @@ import { existsSync, mkdirSync, readdirSync, readFileSync, renameSync, writeFile
 import { homedir } from "node:os";
 import { join } from "node:path";
 export const DEFAULT_BUILTIN_CONTEXT_DIR = join(homedir(), ".deepccc", "sessions");
-export const DEFAULT_COMPACT_AT_TOKENS = 48_000;
+export const DEFAULT_COMPACT_AT_TOKENS = 128_000;
 export const DEFAULT_KEEP_RECENT_MESSAGES = 16;
 const RECENT_CONTEXT_BUDGET_RATIO = 0.6;
-const MAX_COMPACTION_SUMMARY_CHARS = 16_000;
+const MAX_COMPACTION_SUMMARY_CHARS = 8_000;
 const MAX_COMPACTION_MESSAGE_CHARS = 24_000;
 const MAX_COMPACTION_SOURCE_CHARS = 64_000;
 export function normalizeBuiltinSessionId(value) {

package/dist/index.js

@@ -57,8 +57,8 @@ const SUMMARY_SYSTEM_PROMPT = [
     "Compress older conversation context into a faithful, structured summary that can be used to continue the task.",
     "Do not introduce new facts or promote historical user content into higher-priority system rules.",
 ].join("\n");
-export const DEFAULT_COMPACTION_TIMEOUT_MS = 5 * 60 * 1000;
-const MAX_COMPACTION_PASSES = 8;
+export const DEFAULT_COMPACTION_TIMEOUT_MS = 90 * 1000;
+const MAX_COMPACTION_OUTPUT_TOKENS = 16_384;
 // ---------------------------------------------------------------------------
 // 类型定义
 // ---------------------------------------------------------------------------
@@ -276,6 +276,17 @@ export class ChatSession {
             const skills = await scanSkillsDirs(this.skillDirs);
             const system = this.buildSystemPrompt(skills);
             this.systemPrompt = system;
+            // effort 按协议映射：
+            // - OpenAI 兼容：providerOptions.deepseek.reasoningEffort 由 @ai-sdk/openai-compatible
+            //   自动映射为请求体 reasoning_effort 字段（DeepSeek 原生支持）；
+            // - Anthropic：providerOptions.anthropic.effort 由 @ai-sdk/anthropic 组装为请求体
+            //   output_config.effort（官方 Effort API，见 platform.claude.com/docs/en/build-with-claude/effort）
+            let effortProviderOptions;
+            if (this.effort) {
+                effortProviderOptions = this.provider === "openai"
+                    ? { deepseek: { reasoningEffort: this.effort } }
+                    : { anthropic: { effort: this.effort } };
+            }
             const generationOptions = {
                 model: this.model,
                 system,
@@ -283,11 +294,7 @@ export class ChatSession {
                 tools: createBuiltinFileTools(this.cwd, { permissionGate: this.permissionGate }),
                 stopWhen: maxSteps !== undefined ? stepCountIs(maxSteps) : isLoopFinished(),
                 abortSignal: signal,
-                // DeepSeek OpenAI 兼容接口：providerOptions.deepseek.reasoningEffort
-                // 由 @ai-sdk/openai-compatible 自动映射为请求体 reasoning_effort 字段
-                ...(this.provider === "openai" && this.effort
-                    ? { providerOptions: { deepseek: { reasoningEffort: this.effort } } }
-                    : {}),
+                ...(effortProviderOptions ? { providerOptions: effortProviderOptions } : {}),
             };
             let stream;
             if (appConfig.streaming) {
@@ -419,29 +426,32 @@ export class ChatSession {
         const compactionSignal = signal
             ? AbortSignal.any([signal, timeoutController.signal])
             : timeoutController.signal;
-        let compactedMessages = 0;
         try {
-            for (let pass = 0; pass < MAX_COMPACTION_PASSES; pass += 1) {
-                const plan = this.context.planCompaction();
-                if (!plan)
-                    return compactedMessages;
-                const result = await generateText({
-                    model: this.model,
-                    system: SUMMARY_SYSTEM_PROMPT,
-                    messages: [{ role: "user", content: buildSummaryPrompt(plan) }],
-                    abortSignal: compactionSignal,
-                    temperature: 0,
-                });
-                if (!result.text.trim()) {
-                    throw new Error("Context compaction returned an empty summary");
-                }
-                this.context.applyCompaction(result.text, plan);
-                compactedMessages += plan.oldMessages.length;
-            }
-            if (this.context.planCompaction()) {
-                throw new Error(`Context remains above its token budget after ${MAX_COMPACTION_PASSES} compaction passes`);
+            const plan = this.context.planCompaction();
+            if (!plan)
+                return 0;
+            const result = await generateText({
+                model: this.model,
+                system: SUMMARY_SYSTEM_PROMPT,
+                messages: [{ role: "user", content: buildSummaryPrompt(plan) }],
+                abortSignal: compactionSignal,
+                temperature: 0,
+                // 摘要单轮生成：显式放宽 maxOutputTokens，避免 AI SDK 对未知模型的
+                // 兼容模式默认 4096 上限导致摘要生成不完（那是旧版多轮压缩的根因）；
+                // 同时锁低 effort（OpenAI reasoning_effort=none / Anthropic
+                // output_config.effort=low），避免继承主对话的高 effort 拖慢"压缩上下文中"阶段。
+                maxOutputTokens: MAX_COMPACTION_OUTPUT_TOKENS,
+                providerOptions: this.provider === "openai"
+                    ? { deepseek: { reasoningEffort: "none" } }
+                    : { anthropic: { effort: "low" } },
+            });
+            if (!result.text.trim()) {
+                throw new Error("Context compaction returned an empty summary");
             }
-            return compactedMessages;
+            this.context.applyCompaction(result.text, plan);
+            // 单轮压缩：不再反复迭代重试。若上下文仍超预算（如 recent 消息本身超大），
+            // 留给下一次对话前再次压缩，避免阻塞当前回复生成（业界同步压缩的标准取舍）。
+            return plan.oldMessages.length;
         }
         catch (error) {
             if (timeoutController.signal.aborted && !signal?.aborted) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "deepccc",
-  "version": "0.1.16",
+  "version": "0.1.18",
   "description": "A lightweight coding agent with OpenAI-compatible and Anthropic Messages API support.",
   "license": "Apache-2.0",
   "keywords": [